Update README.md

README.md

@@ -44,7 +44,9 @@ def crop_roi(img):
 ```
 
 While not guaranteed to be foolproof, a cursory review of a sample of cropped images demonstrated excellent performance. 
-The model was trained with a large batch size (256) to mitigate noise. 
+The model was trained with a larger batch size (256) to mitigate noise. Input into the model is expected to be [0, 255].
+If you are loading from DICOM, you should convert the input into an 8-bit image, pass it through `model.preprocess`, and
+ensure that it is a float `torch.Tensor`before passing it to the model. The normalization step happens within the model itself.
 
 To use the model:
 ```